2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #define _SVID_SOURCE //needed for MAP_ANONYMOUS
22 #define _DARWIN_C_SOURCE // needed for MAP_ANON
31 #if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
32 #define MAP_ANONYMOUS MAP_ANON
36 #define WIN32_LEAN_AND_MEAN
40 #include "swscale_internal.h"
42 #include "libavutil/intreadwrite.h"
43 #include "libavutil/x86_cpu.h"
44 #include "libavutil/cpu.h"
45 #include "libavutil/avutil.h"
46 #include "libavutil/bswap.h"
47 #include "libavutil/opt.h"
48 #include "libavutil/pixdesc.h"
49 #include "libavutil/avassert.h"
51 unsigned swscale_version(void)
53 return LIBSWSCALE_VERSION_INT;
56 const char *swscale_configuration(void)
58 return FFMPEG_CONFIGURATION;
61 const char *swscale_license(void)
63 #define LICENSE_PREFIX "libswscale license: "
64 return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
67 #define RET 0xC3 //near return opcode for x86
69 typedef struct FormatEntry {
70 int is_supported_in, is_supported_out;
73 const static FormatEntry format_entries[PIX_FMT_NB] = {
74 [PIX_FMT_YUV420P] = { 1 , 1 },
75 [PIX_FMT_YUYV422] = { 1 , 1 },
76 [PIX_FMT_RGB24] = { 1 , 1 },
77 [PIX_FMT_BGR24] = { 1 , 1 },
78 [PIX_FMT_YUV422P] = { 1 , 1 },
79 [PIX_FMT_YUV444P] = { 1 , 1 },
80 [PIX_FMT_YUV410P] = { 1 , 1 },
81 [PIX_FMT_YUV411P] = { 1 , 1 },
82 [PIX_FMT_GRAY8] = { 1 , 1 },
83 [PIX_FMT_MONOWHITE] = { 1 , 1 },
84 [PIX_FMT_MONOBLACK] = { 1 , 1 },
85 [PIX_FMT_PAL8] = { 1 , 0 },
86 [PIX_FMT_YUVJ420P] = { 1 , 1 },
87 [PIX_FMT_YUVJ422P] = { 1 , 1 },
88 [PIX_FMT_YUVJ444P] = { 1 , 1 },
89 [PIX_FMT_UYVY422] = { 1 , 1 },
90 [PIX_FMT_UYYVYY411] = { 0 , 0 },
91 [PIX_FMT_BGR8] = { 1 , 1 },
92 [PIX_FMT_BGR4] = { 0 , 1 },
93 [PIX_FMT_BGR4_BYTE] = { 1 , 1 },
94 [PIX_FMT_RGB8] = { 1 , 1 },
95 [PIX_FMT_RGB4] = { 0 , 1 },
96 [PIX_FMT_RGB4_BYTE] = { 1 , 1 },
97 [PIX_FMT_NV12] = { 1 , 1 },
98 [PIX_FMT_NV21] = { 1 , 1 },
99 [PIX_FMT_ARGB] = { 1 , 1 },
100 [PIX_FMT_RGBA] = { 1 , 1 },
101 [PIX_FMT_ABGR] = { 1 , 1 },
102 [PIX_FMT_BGRA] = { 1 , 1 },
103 [PIX_FMT_GRAY16BE] = { 1 , 1 },
104 [PIX_FMT_GRAY16LE] = { 1 , 1 },
105 [PIX_FMT_YUV440P] = { 1 , 1 },
106 [PIX_FMT_YUVJ440P] = { 1 , 1 },
107 [PIX_FMT_YUVA420P] = { 1 , 1 },
108 [PIX_FMT_RGB48BE] = { 1 , 1 },
109 [PIX_FMT_RGB48LE] = { 1 , 1 },
110 [PIX_FMT_RGB565BE] = { 1 , 1 },
111 [PIX_FMT_RGB565LE] = { 1 , 1 },
112 [PIX_FMT_RGB555BE] = { 1 , 1 },
113 [PIX_FMT_RGB555LE] = { 1 , 1 },
114 [PIX_FMT_BGR565BE] = { 1 , 1 },
115 [PIX_FMT_BGR565LE] = { 1 , 1 },
116 [PIX_FMT_BGR555BE] = { 1 , 1 },
117 [PIX_FMT_BGR555LE] = { 1 , 1 },
118 [PIX_FMT_YUV420P16LE] = { 1 , 1 },
119 [PIX_FMT_YUV420P16BE] = { 1 , 1 },
120 [PIX_FMT_YUV422P16LE] = { 1 , 1 },
121 [PIX_FMT_YUV422P16BE] = { 1 , 1 },
122 [PIX_FMT_YUV444P16LE] = { 1 , 1 },
123 [PIX_FMT_YUV444P16BE] = { 1 , 1 },
124 [PIX_FMT_RGB444LE] = { 0 , 1 },
125 [PIX_FMT_RGB444BE] = { 0 , 1 },
126 [PIX_FMT_BGR444LE] = { 0 , 1 },
127 [PIX_FMT_BGR444BE] = { 0 , 1 },
128 [PIX_FMT_Y400A] = { 1 , 0 },
129 [PIX_FMT_BGR48BE] = { 1 , 1 },
130 [PIX_FMT_BGR48LE] = { 1 , 1 },
131 [PIX_FMT_YUV420P9BE] = { 1 , 1 },
132 [PIX_FMT_YUV420P9LE] = { 1 , 1 },
133 [PIX_FMT_YUV420P10BE] = { 1 , 1 },
134 [PIX_FMT_YUV420P10LE] = { 1 , 1 },
135 [PIX_FMT_YUV422P10BE] = { 1 , 1 },
136 [PIX_FMT_YUV422P10LE] = { 1 , 1 },
137 [PIX_FMT_YUV444P9BE] = { 1 , 0 },
138 [PIX_FMT_YUV444P9LE] = { 1 , 0 },
139 [PIX_FMT_YUV444P10BE] = { 1 , 0 },
140 [PIX_FMT_YUV444P10LE] = { 1 , 0 },
143 int sws_isSupportedInput(enum PixelFormat pix_fmt)
145 return (unsigned)pix_fmt < PIX_FMT_NB ?
146 format_entries[pix_fmt].is_supported_in : 0;
149 int sws_isSupportedOutput(enum PixelFormat pix_fmt)
151 return (unsigned)pix_fmt < PIX_FMT_NB ?
152 format_entries[pix_fmt].is_supported_out : 0;
155 extern const int32_t ff_yuv2rgb_coeffs[8][4];
157 #if FF_API_SWS_FORMAT_NAME
158 const char *sws_format_name(enum PixelFormat format)
160 return av_get_pix_fmt_name(format);
164 static double getSplineCoeff(double a, double b, double c, double d, double dist)
166 if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a;
167 else return getSplineCoeff( 0.0,
174 static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc,
175 int srcW, int dstW, int filterAlign, int one, int flags, int cpu_flags,
176 SwsVector *srcFilter, SwsVector *dstFilter, double param[2])
182 int64_t *filter=NULL;
183 int64_t *filter2=NULL;
184 const int64_t fone= 1LL<<54;
187 emms_c(); //FIXME this should not be required but it IS (even for non-MMX versions)
189 // NOTE: the +3 is for the MMX(+1)/SSE(+3) scaler which reads over the end
190 FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+3)*sizeof(int16_t), fail);
192 if (FFABS(xInc - 0x10000) <10) { // unscaled
195 FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
197 for (i=0; i<dstW; i++) {
198 filter[i*filterSize]= fone;
202 } else if (flags&SWS_POINT) { // lame looking point sampling mode
206 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
208 xDstInSrc= xInc/2 - 0x8000;
209 for (i=0; i<dstW; i++) {
210 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
216 } else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale
220 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
222 xDstInSrc= xInc/2 - 0x8000;
223 for (i=0; i<dstW; i++) {
224 int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16;
228 //bilinear upscale / linear interpolate / area averaging
229 for (j=0; j<filterSize; j++) {
230 int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16);
231 if (coeff<0) coeff=0;
232 filter[i*filterSize + j]= coeff;
241 if (flags&SWS_BICUBIC) sizeFactor= 4;
242 else if (flags&SWS_X) sizeFactor= 8;
243 else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear
244 else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;)
245 else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6;
246 else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;)
247 else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;)
248 else if (flags&SWS_BILINEAR) sizeFactor= 2;
250 sizeFactor= 0; //GCC warning killer
254 if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale
255 else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW;
257 if (filterSize > srcW-2) filterSize=srcW-2;
259 FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail);
261 xDstInSrc= xInc - 0x10000;
262 for (i=0; i<dstW; i++) {
263 int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17);
266 for (j=0; j<filterSize; j++) {
267 int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13;
273 floatd= d * (1.0/(1<<30));
275 if (flags & SWS_BICUBIC) {
276 int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24);
277 int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24);
278 int64_t dd = ( d*d)>>30;
279 int64_t ddd= (dd*d)>>30;
282 coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30);
283 else if (d < 1LL<<31)
284 coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30);
287 coeff *= fone>>(30+24);
289 /* else if (flags & SWS_X) {
290 double p= param ? param*0.01 : 0.3;
291 coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
292 coeff*= pow(2.0, - p*d*d);
294 else if (flags & SWS_X) {
295 double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0;
299 c = cos(floatd*M_PI);
302 if (c<0.0) c= -pow(-c, A);
304 coeff= (c*0.5 + 0.5)*fone;
305 } else if (flags & SWS_AREA) {
306 int64_t d2= d - (1<<29);
307 if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16));
308 else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16));
310 coeff *= fone>>(30+16);
311 } else if (flags & SWS_GAUSS) {
312 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
313 coeff = (pow(2.0, - p*floatd*floatd))*fone;
314 } else if (flags & SWS_SINC) {
315 coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone;
316 } else if (flags & SWS_LANCZOS) {
317 double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0;
318 coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone;
319 if (floatd>p) coeff=0;
320 } else if (flags & SWS_BILINEAR) {
322 if (coeff<0) coeff=0;
324 } else if (flags & SWS_SPLINE) {
325 double p=-2.196152422706632;
326 coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone;
328 coeff= 0.0; //GCC warning killer
332 filter[i*filterSize + j]= coeff;
339 /* apply src & dst Filter to filter -> filter2
342 assert(filterSize>0);
343 filter2Size= filterSize;
344 if (srcFilter) filter2Size+= srcFilter->length - 1;
345 if (dstFilter) filter2Size+= dstFilter->length - 1;
346 assert(filter2Size>0);
347 FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail);
349 for (i=0; i<dstW; i++) {
353 for (k=0; k<srcFilter->length; k++) {
354 for (j=0; j<filterSize; j++)
355 filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j];
358 for (j=0; j<filterSize; j++)
359 filter2[i*filter2Size + j]= filter[i*filterSize + j];
363 (*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2;
367 /* try to reduce the filter-size (step1 find size and shift left) */
368 // Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
370 for (i=dstW-1; i>=0; i--) {
371 int min= filter2Size;
375 /* get rid of near zero elements on the left by shifting left */
376 for (j=0; j<filter2Size; j++) {
378 cutOff += FFABS(filter2[i*filter2Size]);
380 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
382 /* preserve monotonicity because the core can't handle the filter otherwise */
383 if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break;
385 // move filter coefficients left
386 for (k=1; k<filter2Size; k++)
387 filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
388 filter2[i*filter2Size + k - 1]= 0;
393 /* count near zeros on the right */
394 for (j=filter2Size-1; j>0; j--) {
395 cutOff += FFABS(filter2[i*filter2Size + j]);
397 if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break;
401 if (min>minFilterSize) minFilterSize= min;
404 if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) {
405 // we can handle the special case 4,
406 // so we don't want to go to the full 8
407 if (minFilterSize < 5)
410 // We really don't want to waste our time
411 // doing useless computation, so fall back on
412 // the scalar C code for very small filters.
413 // Vectorizing is worth it only if you have a
414 // decent-sized vector.
415 if (minFilterSize < 3)
419 if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) {
420 // special case for unscaled vertical filtering
421 if (minFilterSize == 1 && filterAlign == 2)
425 assert(minFilterSize > 0);
426 filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1));
427 assert(filterSize > 0);
428 filter= av_malloc(filterSize*dstW*sizeof(*filter));
429 if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter)
431 *outFilterSize= filterSize;
433 if (flags&SWS_PRINT_INFO)
434 av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize);
435 /* try to reduce the filter-size (step2 reduce it) */
436 for (i=0; i<dstW; i++) {
439 for (j=0; j<filterSize; j++) {
440 if (j>=filter2Size) filter[i*filterSize + j]= 0;
441 else filter[i*filterSize + j]= filter2[i*filter2Size + j];
442 if((flags & SWS_BITEXACT) && j>=minFilterSize)
443 filter[i*filterSize + j]= 0;
447 //FIXME try to align filterPos if possible
450 for (i=0; i<dstW; i++) {
452 if ((*filterPos)[i] < 0) {
453 // move filter coefficients left to compensate for filterPos
454 for (j=1; j<filterSize; j++) {
455 int left= FFMAX(j + (*filterPos)[i], 0);
456 filter[i*filterSize + left] += filter[i*filterSize + j];
457 filter[i*filterSize + j]=0;
462 if ((*filterPos)[i] + filterSize > srcW) {
463 int shift= (*filterPos)[i] + filterSize - srcW;
464 // move filter coefficients right to compensate for filterPos
465 for (j=filterSize-2; j>=0; j--) {
466 int right= FFMIN(j + shift, filterSize-1);
467 filter[i*filterSize +right] += filter[i*filterSize +j];
468 filter[i*filterSize +j]=0;
470 (*filterPos)[i]= srcW - filterSize;
474 // Note the +1 is for the MMX scaler which reads over the end
475 /* align at 16 for AltiVec (needed by hScale_altivec_real) */
476 FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+3)*sizeof(int16_t), fail);
478 /* normalize & store in outFilter */
479 for (i=0; i<dstW; i++) {
484 for (j=0; j<filterSize; j++) {
485 sum+= filter[i*filterSize + j];
487 sum= (sum + one/2)/ one;
488 for (j=0; j<*outFilterSize; j++) {
489 int64_t v= filter[i*filterSize + j] + error;
490 int intV= ROUNDED_DIV(v, sum);
491 (*outFilter)[i*(*outFilterSize) + j]= intV;
496 (*filterPos)[dstW+0] =
497 (*filterPos)[dstW+1] =
498 (*filterPos)[dstW+2] = (*filterPos)[dstW-1]; // the MMX/SSE scaler will read over the end
499 for (i=0; i<*outFilterSize; i++) {
500 int k= (dstW - 1) * (*outFilterSize) + i;
501 (*outFilter)[k + 1 * (*outFilterSize)] =
502 (*outFilter)[k + 2 * (*outFilterSize)] =
503 (*outFilter)[k + 3 * (*outFilterSize)] = (*outFilter)[k];
514 static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits)
517 x86_reg imm8OfPShufW1A;
518 x86_reg imm8OfPShufW2A;
519 x86_reg fragmentLengthA;
521 x86_reg imm8OfPShufW1B;
522 x86_reg imm8OfPShufW2B;
523 x86_reg fragmentLengthB;
528 // create an optimized horizontal scaling routine
529 /* This scaler is made of runtime-generated MMX2 code using specially
530 * tuned pshufw instructions. For every four output pixels, if four
531 * input pixels are enough for the fast bilinear scaling, then a chunk
532 * of fragmentB is used. If five input pixels are needed, then a chunk
533 * of fragmentA is used.
542 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
543 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
544 "movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t"
545 "punpcklbw %%mm7, %%mm1 \n\t"
546 "punpcklbw %%mm7, %%mm0 \n\t"
547 "pshufw $0xFF, %%mm1, %%mm1 \n\t"
549 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
551 "psubw %%mm1, %%mm0 \n\t"
552 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
553 "pmullw %%mm3, %%mm0 \n\t"
554 "psllw $7, %%mm1 \n\t"
555 "paddw %%mm1, %%mm0 \n\t"
557 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
559 "add $8, %%"REG_a" \n\t"
563 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
564 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
565 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
570 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
574 :"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
575 "=r" (fragmentLengthA)
582 "movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t"
583 "movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t"
584 "punpcklbw %%mm7, %%mm0 \n\t"
585 "pshufw $0xFF, %%mm0, %%mm1 \n\t"
587 "pshufw $0xFF, %%mm0, %%mm0 \n\t"
589 "psubw %%mm1, %%mm0 \n\t"
590 "movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t"
591 "pmullw %%mm3, %%mm0 \n\t"
592 "psllw $7, %%mm1 \n\t"
593 "paddw %%mm1, %%mm0 \n\t"
595 "movq %%mm0, (%%"REG_D", %%"REG_a") \n\t"
597 "add $8, %%"REG_a" \n\t"
601 "lea " LOCAL_MANGLE(0b) ", %0 \n\t"
602 "lea " LOCAL_MANGLE(1b) ", %1 \n\t"
603 "lea " LOCAL_MANGLE(2b) ", %2 \n\t"
608 "lea " LOCAL_MANGLE(9b) ", %3 \n\t"
612 :"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
613 "=r" (fragmentLengthB)
616 xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
619 for (i=0; i<dstW/numSplits; i++) {
624 int b=((xpos+xInc)>>16) - xx;
625 int c=((xpos+xInc*2)>>16) - xx;
626 int d=((xpos+xInc*3)>>16) - xx;
628 uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA;
629 x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A;
630 x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A;
631 x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA;
632 int maxShift= 3-(d+inc);
636 filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9;
637 filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9;
638 filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
639 filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
642 memcpy(filterCode + fragmentPos, fragment, fragmentLength);
644 filterCode[fragmentPos + imm8OfPShufW1]=
645 (a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6);
646 filterCode[fragmentPos + imm8OfPShufW2]=
647 a | (b<<2) | (c<<4) | (d<<6);
649 if (i+4-inc>=dstW) shift=maxShift; //avoid overread
650 else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align
652 if (shift && i>=shift) {
653 filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
654 filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
655 filterPos[i/2]-=shift;
659 fragmentPos+= fragmentLength;
662 filterCode[fragmentPos]= RET;
667 filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part
669 return fragmentPos + 1;
671 #endif /* HAVE_MMX2 */
673 static void getSubSampleFactors(int *h, int *v, enum PixelFormat format)
675 *h = av_pix_fmt_descriptors[format].log2_chroma_w;
676 *v = av_pix_fmt_descriptors[format].log2_chroma_h;
679 int sws_setColorspaceDetails(struct SwsContext *c, const int inv_table[4],
680 int srcRange, const int table[4], int dstRange,
681 int brightness, int contrast, int saturation)
683 memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4);
684 memcpy(c->dstColorspaceTable, table, sizeof(int)*4);
686 c->brightness= brightness;
687 c->contrast = contrast;
688 c->saturation= saturation;
689 c->srcRange = srcRange;
690 c->dstRange = dstRange;
691 if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
693 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]);
694 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]);
696 ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation);
699 if (HAVE_ALTIVEC && av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)
700 ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation);
704 int sws_getColorspaceDetails(struct SwsContext *c, int **inv_table,
705 int *srcRange, int **table, int *dstRange,
706 int *brightness, int *contrast, int *saturation)
708 if (!c || isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1;
710 *inv_table = c->srcColorspaceTable;
711 *table = c->dstColorspaceTable;
712 *srcRange = c->srcRange;
713 *dstRange = c->dstRange;
714 *brightness= c->brightness;
715 *contrast = c->contrast;
716 *saturation= c->saturation;
721 static int handle_jpeg(enum PixelFormat *format)
724 case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1;
725 case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1;
726 case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1;
727 case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1;
732 SwsContext *sws_alloc_context(void)
734 SwsContext *c= av_mallocz(sizeof(SwsContext));
736 c->av_class = &sws_context_class;
737 av_opt_set_defaults(c);
742 int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
745 int usesVFilter, usesHFilter;
747 SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
752 int dst_stride = FFALIGN(dstW * sizeof(int16_t)+66, 16);
753 int flags, cpu_flags;
754 enum PixelFormat srcFormat= c->srcFormat;
755 enum PixelFormat dstFormat= c->dstFormat;
757 cpu_flags = av_get_cpu_flags();
760 if (!rgb15to16) sws_rgb2rgb_init();
762 unscaled = (srcW == dstW && srcH == dstH);
764 if (!sws_isSupportedInput(srcFormat)) {
765 av_log(c, AV_LOG_ERROR, "%s is not supported as input pixel format\n", av_get_pix_fmt_name(srcFormat));
766 return AVERROR(EINVAL);
768 if (!sws_isSupportedOutput(dstFormat)) {
769 av_log(c, AV_LOG_ERROR, "%s is not supported as output pixel format\n", av_get_pix_fmt_name(dstFormat));
770 return AVERROR(EINVAL);
773 i= flags & ( SWS_POINT
784 if(!i || (i & (i-1))) {
785 av_log(c, AV_LOG_ERROR, "Exactly one scaler algorithm must be chosen\n");
786 return AVERROR(EINVAL);
789 if (srcW<4 || srcH<1 || dstW<8 || dstH<1) { //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code
790 av_log(c, AV_LOG_ERROR, "%dx%d -> %dx%d is invalid scaling dimension\n",
791 srcW, srcH, dstW, dstH);
792 return AVERROR(EINVAL);
795 if (!dstFilter) dstFilter= &dummyFilter;
796 if (!srcFilter) srcFilter= &dummyFilter;
798 c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW;
799 c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH;
800 c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]);
801 c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]);
802 c->vRounder= 4* 0x0001000100010001ULL;
804 usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
805 (srcFilter->chrV && srcFilter->chrV->length>1) ||
806 (dstFilter->lumV && dstFilter->lumV->length>1) ||
807 (dstFilter->chrV && dstFilter->chrV->length>1);
808 usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
809 (srcFilter->chrH && srcFilter->chrH->length>1) ||
810 (dstFilter->lumH && dstFilter->lumH->length>1) ||
811 (dstFilter->chrH && dstFilter->chrH->length>1);
813 getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat);
814 getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat);
816 // reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
817 if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1;
819 // drop some chroma lines if the user wants it
820 c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT;
821 c->chrSrcVSubSample+= c->vChrDrop;
823 // drop every other pixel for chroma calculation unless user wants full chroma
824 if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
825 && srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8
826 && srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4
827 && srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE
828 && ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
829 c->chrSrcHSubSample=1;
831 // Note the -((-x)>>y) is so that we always round toward +inf.
832 c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample);
833 c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample);
834 c->chrDstW= -((-dstW) >> c->chrDstHSubSample);
835 c->chrDstH= -((-dstH) >> c->chrDstVSubSample);
837 /* unscaled special cases */
838 if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
839 ff_get_unscaled_swscale(c);
842 if (flags&SWS_PRINT_INFO)
843 av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
844 av_get_pix_fmt_name(srcFormat), av_get_pix_fmt_name(dstFormat));
849 c->srcBpc = 1 + av_pix_fmt_descriptors[srcFormat].comp[0].depth_minus1;
852 c->dstBpc = 1 + av_pix_fmt_descriptors[dstFormat].comp[0].depth_minus1;
855 if (isAnyRGB(srcFormat) || srcFormat == PIX_FMT_PAL8)
859 FF_ALLOC_OR_GOTO(c, c->formatConvBuffer, FFALIGN(srcW*2+78, 16) * 2, fail);
860 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2 && c->srcBpc == 8 && c->dstBpc <= 10) {
861 c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0;
862 if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) {
863 if (flags&SWS_PRINT_INFO)
864 av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
866 if (usesHFilter || isNBPS(c->srcFormat) || is16BPS(c->srcFormat) || isAnyRGB(c->srcFormat)) c->canMMX2BeUsed=0;
871 c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW;
872 c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH;
874 // match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
875 // but only for the FAST_BILINEAR mode otherwise do correct scaling
876 // n-2 is the last chrominance sample available
877 // this is not perfect, but no one should notice the difference, the more correct variant
878 // would be like the vertical one, but that would require some special code for the
879 // first and last pixel
880 if (flags&SWS_FAST_BILINEAR) {
881 if (c->canMMX2BeUsed) {
885 //we don't use the x86 asm scaler if MMX is available
886 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX && c->dstBpc <= 10) {
887 c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20;
888 c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20;
892 /* precalculate horizontal scaler filter coefficients */
895 // can't downscale !!!
896 if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
897 c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8);
898 c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4);
901 c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
902 c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
903 #elif HAVE_VIRTUALALLOC
904 c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
905 c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
907 c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize);
908 c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize);
912 if (c->lumMmx2FilterCode == MAP_FAILED || c->chrMmx2FilterCode == MAP_FAILED)
914 if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
916 return AVERROR(ENOMEM);
917 FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail);
918 FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail);
919 FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail);
920 FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail);
922 initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
923 initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
926 mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
927 mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ);
930 #endif /* HAVE_MMX2 */
932 const int filterAlign=
933 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) ? 4 :
934 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
937 if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
938 srcW , dstW, filterAlign, 1<<14,
939 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
940 srcFilter->lumH, dstFilter->lumH, c->param) < 0)
942 if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
943 c->chrSrcW, c->chrDstW, filterAlign, 1<<14,
944 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
945 srcFilter->chrH, dstFilter->chrH, c->param) < 0)
948 } // initialize horizontal stuff
950 /* precalculate vertical scaler filter coefficients */
952 const int filterAlign=
953 (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
954 (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) ? 8 :
957 if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
958 srcH , dstH, filterAlign, (1<<12),
959 (flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, cpu_flags,
960 srcFilter->lumV, dstFilter->lumV, c->param) < 0)
962 if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
963 c->chrSrcH, c->chrDstH, filterAlign, (1<<12),
964 (flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, cpu_flags,
965 srcFilter->chrV, dstFilter->chrV, c->param) < 0)
969 FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail);
970 FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail);
972 for (i=0;i<c->vLumFilterSize*c->dstH;i++) {
974 short *p = (short *)&c->vYCoeffsBank[i];
976 p[j] = c->vLumFilter[i];
979 for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) {
981 short *p = (short *)&c->vCCoeffsBank[i];
983 p[j] = c->vChrFilter[i];
988 // calculate buffer sizes so that they won't run out while handling these damn slices
989 c->vLumBufSize= c->vLumFilterSize;
990 c->vChrBufSize= c->vChrFilterSize;
991 for (i=0; i<dstH; i++) {
992 int chrI= (int64_t)i*c->chrDstH / dstH;
993 int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1,
994 ((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
996 nextSlice>>= c->chrSrcVSubSample;
997 nextSlice<<= c->chrSrcVSubSample;
998 if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
999 c->vLumBufSize= nextSlice - c->vLumFilterPos[i];
1000 if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
1001 c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI];
1004 // allocate pixbufs (we use dynamic allocation because otherwise we would need to
1005 // allocate several megabytes to handle all possible cases)
1006 FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1007 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1008 FF_ALLOC_OR_GOTO(c, c->chrVPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail);
1009 if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
1010 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail);
1011 //Note we need at least one pixel more at the end because of the MMX code (just in case someone wanna replace the 4000/8000)
1012 /* align at 16 bytes for AltiVec */
1013 for (i=0; i<c->vLumBufSize; i++) {
1014 FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1015 c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize];
1017 // 64 / c->scalingBpp is the same as 16 / sizeof(scaling_intermediate)
1018 c->uv_off = (dst_stride>>1) + 64 / (c->dstBpc &~ 7);
1019 c->uv_offx2 = dst_stride + 16;
1020 for (i=0; i<c->vChrBufSize; i++) {
1021 FF_ALLOC_OR_GOTO(c, c->chrUPixBuf[i+c->vChrBufSize], dst_stride*2+32, fail);
1022 c->chrUPixBuf[i] = c->chrUPixBuf[i+c->vChrBufSize];
1023 c->chrVPixBuf[i] = c->chrVPixBuf[i+c->vChrBufSize] = c->chrUPixBuf[i] + (dst_stride >> 1) + 8;
1025 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
1026 for (i=0; i<c->vLumBufSize; i++) {
1027 FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], dst_stride+16, fail);
1028 c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize];
1031 //try to avoid drawing green stuff between the right end and the stride end
1032 for (i=0; i<c->vChrBufSize; i++)
1033 if(av_pix_fmt_descriptors[c->dstFormat].comp[0].depth_minus1 == 15){
1034 av_assert0(c->dstBpc > 10);
1035 for(j=0; j<dst_stride/2+1; j++)
1036 ((int32_t*)(c->chrUPixBuf[i]))[j] = 1<<18;
1038 for(j=0; j<dst_stride+1; j++)
1039 ((int16_t*)(c->chrUPixBuf[i]))[j] = 1<<14;
1041 assert(c->chrDstH <= dstH);
1043 if (flags&SWS_PRINT_INFO) {
1044 if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, ");
1045 else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, ");
1046 else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, ");
1047 else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, ");
1048 else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, ");
1049 else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, ");
1050 else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, ");
1051 else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, ");
1052 else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, ");
1053 else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, ");
1054 else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, ");
1055 else av_log(c, AV_LOG_INFO, "ehh flags invalid?! ");
1057 av_log(c, AV_LOG_INFO, "from %s to %s%s ",
1058 av_get_pix_fmt_name(srcFormat),
1060 dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 ||
1061 dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE ||
1062 dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "",
1066 av_get_pix_fmt_name(dstFormat));
1068 if (HAVE_MMX2 && cpu_flags & AV_CPU_FLAG_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n");
1069 else if (HAVE_AMD3DNOW && cpu_flags & AV_CPU_FLAG_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n");
1070 else if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_MMX) av_log(c, AV_LOG_INFO, "using MMX\n");
1071 else if (HAVE_ALTIVEC && cpu_flags & AV_CPU_FLAG_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n");
1072 else av_log(c, AV_LOG_INFO, "using C\n");
1074 av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
1075 av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1076 c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc);
1077 av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
1078 c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc);
1081 c->swScale= ff_getSwsFunc(c);
1083 fail: //FIXME replace things by appropriate error codes
1087 #if FF_API_SWS_GETCONTEXT
1088 SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat,
1089 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1090 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1094 if(!(c=sws_alloc_context()))
1102 c->srcRange = handle_jpeg(&srcFormat);
1103 c->dstRange = handle_jpeg(&dstFormat);
1104 c->srcFormat= srcFormat;
1105 c->dstFormat= dstFormat;
1108 c->param[0] = param[0];
1109 c->param[1] = param[1];
1111 sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, c->dstRange, 0, 1<<16, 1<<16);
1113 if(sws_init_context(c, srcFilter, dstFilter) < 0){
1122 SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur,
1123 float lumaSharpen, float chromaSharpen,
1124 float chromaHShift, float chromaVShift,
1127 SwsFilter *filter= av_malloc(sizeof(SwsFilter));
1131 if (lumaGBlur!=0.0) {
1132 filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0);
1133 filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0);
1135 filter->lumH= sws_getIdentityVec();
1136 filter->lumV= sws_getIdentityVec();
1139 if (chromaGBlur!=0.0) {
1140 filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0);
1141 filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0);
1143 filter->chrH= sws_getIdentityVec();
1144 filter->chrV= sws_getIdentityVec();
1147 if (chromaSharpen!=0.0) {
1148 SwsVector *id= sws_getIdentityVec();
1149 sws_scaleVec(filter->chrH, -chromaSharpen);
1150 sws_scaleVec(filter->chrV, -chromaSharpen);
1151 sws_addVec(filter->chrH, id);
1152 sws_addVec(filter->chrV, id);
1156 if (lumaSharpen!=0.0) {
1157 SwsVector *id= sws_getIdentityVec();
1158 sws_scaleVec(filter->lumH, -lumaSharpen);
1159 sws_scaleVec(filter->lumV, -lumaSharpen);
1160 sws_addVec(filter->lumH, id);
1161 sws_addVec(filter->lumV, id);
1165 if (chromaHShift != 0.0)
1166 sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5));
1168 if (chromaVShift != 0.0)
1169 sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5));
1171 sws_normalizeVec(filter->chrH, 1.0);
1172 sws_normalizeVec(filter->chrV, 1.0);
1173 sws_normalizeVec(filter->lumH, 1.0);
1174 sws_normalizeVec(filter->lumV, 1.0);
1176 if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG);
1177 if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG);
1182 SwsVector *sws_allocVec(int length)
1184 SwsVector *vec = av_malloc(sizeof(SwsVector));
1187 vec->length = length;
1188 vec->coeff = av_malloc(sizeof(double) * length);
1194 SwsVector *sws_getGaussianVec(double variance, double quality)
1196 const int length= (int)(variance*quality + 0.5) | 1;
1198 double middle= (length-1)*0.5;
1199 SwsVector *vec= sws_allocVec(length);
1204 for (i=0; i<length; i++) {
1205 double dist= i-middle;
1206 vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI);
1209 sws_normalizeVec(vec, 1.0);
1214 SwsVector *sws_getConstVec(double c, int length)
1217 SwsVector *vec= sws_allocVec(length);
1222 for (i=0; i<length; i++)
1228 SwsVector *sws_getIdentityVec(void)
1230 return sws_getConstVec(1.0, 1);
1233 static double sws_dcVec(SwsVector *a)
1238 for (i=0; i<a->length; i++)
1244 void sws_scaleVec(SwsVector *a, double scalar)
1248 for (i=0; i<a->length; i++)
1249 a->coeff[i]*= scalar;
1252 void sws_normalizeVec(SwsVector *a, double height)
1254 sws_scaleVec(a, height/sws_dcVec(a));
1257 static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
1259 int length= a->length + b->length - 1;
1261 SwsVector *vec= sws_getConstVec(0.0, length);
1266 for (i=0; i<a->length; i++) {
1267 for (j=0; j<b->length; j++) {
1268 vec->coeff[i+j]+= a->coeff[i]*b->coeff[j];
1275 static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
1277 int length= FFMAX(a->length, b->length);
1279 SwsVector *vec= sws_getConstVec(0.0, length);
1284 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1285 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i];
1290 static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
1292 int length= FFMAX(a->length, b->length);
1294 SwsVector *vec= sws_getConstVec(0.0, length);
1299 for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i];
1300 for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i];
1305 /* shift left / or right if "shift" is negative */
1306 static SwsVector *sws_getShiftedVec(SwsVector *a, int shift)
1308 int length= a->length + FFABS(shift)*2;
1310 SwsVector *vec= sws_getConstVec(0.0, length);
1315 for (i=0; i<a->length; i++) {
1316 vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i];
1322 void sws_shiftVec(SwsVector *a, int shift)
1324 SwsVector *shifted= sws_getShiftedVec(a, shift);
1326 a->coeff= shifted->coeff;
1327 a->length= shifted->length;
1331 void sws_addVec(SwsVector *a, SwsVector *b)
1333 SwsVector *sum= sws_sumVec(a, b);
1335 a->coeff= sum->coeff;
1336 a->length= sum->length;
1340 void sws_subVec(SwsVector *a, SwsVector *b)
1342 SwsVector *diff= sws_diffVec(a, b);
1344 a->coeff= diff->coeff;
1345 a->length= diff->length;
1349 void sws_convVec(SwsVector *a, SwsVector *b)
1351 SwsVector *conv= sws_getConvVec(a, b);
1353 a->coeff= conv->coeff;
1354 a->length= conv->length;
1358 SwsVector *sws_cloneVec(SwsVector *a)
1361 SwsVector *vec= sws_allocVec(a->length);
1366 for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i];
1371 void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level)
1378 for (i=0; i<a->length; i++)
1379 if (a->coeff[i]>max) max= a->coeff[i];
1381 for (i=0; i<a->length; i++)
1382 if (a->coeff[i]<min) min= a->coeff[i];
1386 for (i=0; i<a->length; i++) {
1387 int x= (int)((a->coeff[i]-min)*60.0/range +0.5);
1388 av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
1389 for (;x>0; x--) av_log(log_ctx, log_level, " ");
1390 av_log(log_ctx, log_level, "|\n");
1394 void sws_freeVec(SwsVector *a)
1397 av_freep(&a->coeff);
1402 void sws_freeFilter(SwsFilter *filter)
1404 if (!filter) return;
1406 if (filter->lumH) sws_freeVec(filter->lumH);
1407 if (filter->lumV) sws_freeVec(filter->lumV);
1408 if (filter->chrH) sws_freeVec(filter->chrH);
1409 if (filter->chrV) sws_freeVec(filter->chrV);
1413 void sws_freeContext(SwsContext *c)
1419 for (i=0; i<c->vLumBufSize; i++)
1420 av_freep(&c->lumPixBuf[i]);
1421 av_freep(&c->lumPixBuf);
1424 if (c->chrUPixBuf) {
1425 for (i=0; i<c->vChrBufSize; i++)
1426 av_freep(&c->chrUPixBuf[i]);
1427 av_freep(&c->chrUPixBuf);
1428 av_freep(&c->chrVPixBuf);
1431 if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
1432 for (i=0; i<c->vLumBufSize; i++)
1433 av_freep(&c->alpPixBuf[i]);
1434 av_freep(&c->alpPixBuf);
1437 av_freep(&c->vLumFilter);
1438 av_freep(&c->vChrFilter);
1439 av_freep(&c->hLumFilter);
1440 av_freep(&c->hChrFilter);
1442 av_freep(&c->vYCoeffsBank);
1443 av_freep(&c->vCCoeffsBank);
1446 av_freep(&c->vLumFilterPos);
1447 av_freep(&c->vChrFilterPos);
1448 av_freep(&c->hLumFilterPos);
1449 av_freep(&c->hChrFilterPos);
1452 #ifdef MAP_ANONYMOUS
1453 if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
1454 if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
1455 #elif HAVE_VIRTUALALLOC
1456 if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE);
1457 if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE);
1459 av_free(c->lumMmx2FilterCode);
1460 av_free(c->chrMmx2FilterCode);
1462 c->lumMmx2FilterCode=NULL;
1463 c->chrMmx2FilterCode=NULL;
1464 #endif /* HAVE_MMX */
1466 av_freep(&c->yuvTable);
1467 av_freep(&c->formatConvBuffer);
1472 struct SwsContext *sws_getCachedContext(struct SwsContext *context,
1473 int srcW, int srcH, enum PixelFormat srcFormat,
1474 int dstW, int dstH, enum PixelFormat dstFormat, int flags,
1475 SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param)
1477 static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT};
1480 param = default_param;
1483 (context->srcW != srcW ||
1484 context->srcH != srcH ||
1485 context->srcFormat != srcFormat ||
1486 context->dstW != dstW ||
1487 context->dstH != dstH ||
1488 context->dstFormat != dstFormat ||
1489 context->flags != flags ||
1490 context->param[0] != param[0] ||
1491 context->param[1] != param[1])) {
1492 sws_freeContext(context);
1497 if (!(context = sws_alloc_context()))
1499 context->srcW = srcW;
1500 context->srcH = srcH;
1501 context->srcRange = handle_jpeg(&srcFormat);
1502 context->srcFormat = srcFormat;
1503 context->dstW = dstW;
1504 context->dstH = dstH;
1505 context->dstRange = handle_jpeg(&dstFormat);
1506 context->dstFormat = dstFormat;
1507 context->flags = flags;
1508 context->param[0] = param[0];
1509 context->param[1] = param[1];
1510 sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], context->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, context->dstRange, 0, 1<<16, 1<<16);
1511 if (sws_init_context(context, srcFilter, dstFilter) < 0) {
1512 sws_freeContext(context);